A CROSS DISCIPLINARY SURVEY C4 0 Bonnie Lynn Nash-Webber A] R T Bolt Beranek and Newman Inc

I L L N 0 I S UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN PRODUCTION NOTE University of Illinois at Urbana-Champaign Library Large-scale Digitization Project, 2007. 34W~ /52 T E C H] R N 1 E Technical Report No. 31 I] P ANAPHORA: A CROSS DISCIPLINARY SURVEY C4 0 Bonnie Lynn Nash-Webber A] R T Bolt Beranek and Newman Inc. S April 1977 Center for the Study of Reading Th::-E LIBRARY OF THE- 0CT ? 7981 AT UP .. 'PAIGN UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN BOLT BERANEK AND NEWMAN INC. 1005 West Nevada Street 50 Moulton Street Urbana, Illinois 61801 Cambridge, Massachusetts 02138 CENTER FOR THE STUDY OF READING Technical Report No. 31 ANAPHORA: A CROSS DISCIPLINARY SURVEY Bonnie Lynn Nash-Webber Bolt Beranek and Newman Inc. April 1977 BBN Report No. 3546 University of Illinois at Urbana-Champaign Bolt Beranek and Newman Inc. 1005 West Nevada Street 50 Moulton Street Urbana, Illinois 61801 Cambridge, Massachusetts 02138 The research reported herein was supported in part by the National Institute of Education under Contract No. MS-NIE-C- 400-76-0116. A version of this paper will appear in R. J. Spiro, B. C. Bruce, and W. F. Brewer (Eds.), Theoretical Issues in Reading Comprehension, Hillsdale, N.J.: Lawrence Erlbaum Associates. Anaphora Anaphora: A Cross Disciplinary Survey Bonnie Lynn Nash-Webber Bolt Beranek and Newman Inc. Cambridge MA 02138 Abstract Two fundamental assumptions guide this survey of recent research on anaphora. The first is that anaphoric expressions do not refer to segments in a text or discourse, but to entities which are assumed to be in the language receiver's mind. The second assumption is that a text serves to suggest the referents for anaphora, as does the non-linguistic context. As a result, this survey is organized into a consideration of the following questions: 1. What types of entities are possible antecedents in English? 2. What is the relation of the text to these antecedents? 3. What is the relation of inference to these antecedents? 4. What does anaphora say about memory organization? 5. What factors have been posed as influencing anaphor resolution? 6. What is known about anaphora and language acquisition? - 1 - Anaphora Introduction In understanding language understanding, and in particular, reading comprehension, it is almost a truism to say that one cannot stop at the analysis of single sentences alone. In comprehending text, the import of each successive sentence must be determined within, and integrated into, an incrementally growing model of text content and purpose. Now it is also almost a truism that much has been gained already through the formal analyses of sentence-level syntax and semantics that have been put forth in both the context of transformational grammar and that of machine-based natural language question-answering systems (Landsbergen, 1976; Scha, 1976; Winograd, 1972; Woods et al., 1972) In moving from single sentences to connected text, we need to enlarge our domain of analysis. On what does the connectivity of text draw? What inter-sentential devices carry over to text that most important function of sentence-level syntax which Huggins (in press) describes as "a way of maximizing the rate of transfer of meaning from a language producer (a speaker or writer) to a language receiver (a listener or reader), taking into account the limitations of memory of the receiver". What sorts of knowledge and processing heuristics must be possessed by the language receiver to handle text containing such devices? What would result from their absence? One such inter-sentential device, anaphora, is the subject of this survey. Anaphoric expressions comprise pronouns, - 2 - Anaohor a pro-verbs, some definite noun phrases and ellipses. They epitomize a device for "maximizing the rate of transfer of meaning": for example, one short syllable, "it", has the potential for evoking in the language receiver's mind a comolex theoretical construct or an entire chain of events leadina to some conclusion, "It was christened by Feynmann 'the eight-fold way'." "In the end, it drove Lear mad." There are two fundamental assumptions about anaphora in which this survey is grounded. The first is that ananhoric expressions do not refer to segments in a text, but to entities which are assumed to be in the language receiver's mind. The second assumption is that a text serves to suggest the referents for anaphora, as does the non-linguistic context. (The latter is discussed at length in an excellent paper by Hankamer and Saq (1976). It will receive only brief mention here.) The result is a model of comprehension in which the relation of antecedent to anaphor is indirect: the text or non-linguistic environment evokes entities in the language receiver's mind which may be addressed, in turn, anaphorically. (A line is being drawn here between the two notions of anaphora and deixis (Rubin, in press ). Deixis, as another linguistic device for pointing to things, shares with anaphora the above-mentioned function of allowing a language producer to maximize the rate of information flow out to a language receiver. However, deictic expressions are seen as pointing to things within the shared spatial and/or temporal context of language - 3 - Anaphora producer and receiver, while anaphoric expressions are seen as pointing to entities in the language receiver's mind. An effect of deictic pointing to "external" things - "You see that chair there?" - may be to engender "internal" entities which may then be addressed anaphorically - "Well, I paid almost "200 for it.") I see several reasons for discussing anaphora here as an illustration of inter-sentential devices in reading comprehension. First, if a reader does not recognize an expression as anaphoric, or if he or she is unable to handle it as the writer intended, then there is no way that he or she can build up a correct model of the text. Secondly, as recent research in artificial intelligence, psychology, philosophy and linguistics has shown, the process of anaphor resolution may demand very sophisticated syntactic, semantic, pragmatic, inferential and evaluative abilities on the reader's part. Such abilities are even needed to determine what the possible antecedents could be! One might suspect therefore that anaphora might easily be a source of comprehension difficulties. Thirdly, research on anaphora has been very piece-meal (and in rather small pieces, at that) and its observations and results lie scattered through the linguistics, psychology, artificial intelligence and philosophy literatures. There has been no unifying characterization or study of anaphora, and as a result, it is a poorly understood phenomenon. This survey of recent research on anaphora is directed at such a characterization. There is at least one major topic relevant to anaphora which space limitations preclude my taking up here. That is a survey of - 4 - Anaphora computer-based attempts to handle anaphoric expressions. Such attempts may be found in (Baranofsky, 1970; Burton, 1976; Charniak, 1972, 1973; Deutsch, 1975a, 1975b; Grosz, forthcoming; Hobbs, 1976; Klappholz & Lockman, 1975; Levin, 175; Norman & Rumelhart, 1975; Rieger, 1974; Rosenberg, 1976; Wilks, 1975; Winograd, 1972; Woods et al., 1972). While these systems are only first- or second-order attempts at modeling anaphoric processing, they do point to real problems that any more sopfisticated model must overcome. Finally, before I begin, there is one point I would like to emphasize. The formal view of language which guides much of this survey on anaphora is not only compatible with other more pragmatic, intention- or belief-oriented points of view (Morgan, in press) but is entirely complementary. To see this, consider the following example: when asked to recommend John Smith for a vacant assistant professorship, his advisor writes "Mr. Smith has a lovely wife". Viewing this sentence pragmatically will assign it an import which damns Mr. Smith with irrelevant praise. On the other hand, viewing it formally will identify those entities that the sentence evokes in the language receiver's mind: namely, the individuals John and John's wife, the description "lovely wife" and "wife", and the predicate "having a lovely wife". All and only these entities are accessible pronominally or elliptically in subsequent sentences, which may of course continue to reflect the writer's beliefs about John Smith. E.g. Mr. Smith has a lovely wife. - 5 - Ananhora - Moreover, her father attended this university. or - Moreover, his brother does too. or - His previous one was quite ugly. A. Antecedents 1. What types of entities are possible antecedents in English? Probably the most important thing to understand about antecedents is that they are not the elements in the text but are those suggested by it, those concepts being evoked or constructed in the reader's mind. That is, the antecedent of "it" in la. Mary gave Sue a T-shirt. b. She thanked her for it. is not the string "a T-shirt" but the concept the reader should have of the T-shirt that Mary just gave Sue. To some, this may be an obvious point, but the popular misconception that "a pronoun stands for a noun" indicates that it is not as obvious as one might think. Given this observation, the real question becomes: which concepts that should be evoked or built in the reader's head does the language allow one to reference or re-use? Obviously, not everything is a possible antecedent in English: there is, for example, no anaphor whose antecedent is an adjective, a string of adjectives, an adverb, a preposition, or a quantifier. For example, there is no way to get around saying "all except three" in 2a. All except three boys love their mothers. b. _____ girls do too. Researchers have noted many different types of antecedents that English allows.

Load more